home <- here::here()
library(INLA)
library(tidyverse)
library(tidylog)
library(RCurl)
library(sdmTMB)
library(RColorBrewer)
library(devtools)
library(patchwork)
library(ggstats)
library(ggh4x)
library(sdmTMBextra)
# Source map-plot
source_url("https://raw.githubusercontent.com/maxlindmark/cod-interactions/main/R/functions/map-plot.R")
#source(paste0(home, "/R/functions/map-plot.R"))Food competition analysis
Load packages
Read data & prepare data
d <- read_csv(paste0(home, "/data/clean/aggregated_stomach_data.csv"))
# Calculate relative prey weights (saduria and benthos)
d <- d %>%
drop_na(group) %>%
drop_na(oxy) %>%
rename(oxygen = oxy) %>%
mutate(tot_weight = rowSums(select(., ends_with('_tot'))),
benthic_weight = amphipoda_tot + bivalvia_tot + gadus_morhua_tot +
gobiidae_tot + mysidae_tot + non_bio_tot +
other_crustacea_tot + other_tot + other_pisces_tot + platichthys_flesus_tot +
polychaeta_tot + saduria_entomon_tot) %>%
rename(saduria_weight = saduria_entomon_tot,
flounder_density = fle_kg_km2,
large_cod_density = mcod_kg_km2,
small_cod_density = scod_kg_km2) %>%
mutate(tot_rel_weight = tot_weight / (pred_weight_g - tot_weight),
benthic_rel_weight = benthic_weight / (pred_weight_g - tot_weight),
saduria_rel_weight = saduria_weight / (pred_weight_g - tot_weight)) %>%
dplyr::select(-ends_with("_tot")) %>%
dplyr::select(-predator_latin_name, date) %>%
# scale variables
mutate(fyear = as.factor(year),
fquarter = as.factor(quarter),
fhaul_id = as.factor(haul_id),
depth_sc = as.numeric(scale(depth)),
oxygen_sc = as.numeric(scale(oxygen)),
density_saduria_sc = as.numeric(scale(density_saduria)),
flounder_density_sc = as.numeric(scale(flounder_density)),
large_cod_density_sc = as.numeric(scale(large_cod_density)),
small_cod_density_sc = as.numeric(scale(small_cod_density)))Quick explore
Sample size
d %>%
group_by(species) %>%
summarise(n = n())group_by: one grouping variable (species)summarise: now 2 rows and 2 columns, ungrouped# A tibble: 2 × 2
species n
<chr> <int>
1 Cod 5259
2 Flounder 3851d %>%
group_by(species, quarter) %>%
summarise(n = n())group_by: 2 grouping variables (species, quarter)summarise: now 4 rows and 3 columns, one group variable remaining (species)# A tibble: 4 × 3
# Groups: species [2]
species quarter n
<chr> <dbl> <int>
1 Cod 1 3012
2 Cod 4 2247
3 Flounder 1 2081
4 Flounder 4 1770Fit models
Groups are: small cod, large cod and flounder. Response variables are: saduria_rel_weight, benthic_rel_weight or total weight. The latter is only for adult cod, because essentially all prey are benthic for small cod and flounder.
# This is the reason we don't do total weight for flounder and small cod
d %>%
filter(tot_rel_weight > 0) %>%
group_by(group) %>%
mutate(ben_prop = benthic_rel_weight / tot_rel_weight) %>%
summarise(mean_ben_prop = mean(ben_prop))# A tibble: 3 × 2
group mean_ben_prop
<chr> <dbl>
1 flounder 0.978
2 large cod 0.592
3 small cod 0.956Covariates are: ~ 0 + fyear + fquarter + depth_sc + spatial + spatiotemporal random fields + density covariates. For saduria, we use saduria also in interaction with cod and flounder. For cod we use small cod because large and small cod are very correlated. For benthic and total prey, we instead use oxygen, more as a proxy, as the interaction variable
Compare models with different spatial terms and evaluate AIC
- Update, I now do 5-fold cross validation to select models, see script food_competition_crossvalidation.qmd
Main models
pred_flounder_sad <- list()
pred_flounder_ben <- list()
pred_cod_sad <- list()
pred_cod_ben <- list()
coef_sad <- list()
coef_ben <- list()
res_sad <- list()
res_ben <- list()
random_sad <- list()
random_ben <- list()
range_sad <- list()
range_ben <- list()
for(i in unique(d$group)) {
dd <- filter(d, group == i)
mesh <- make_mesh(dd,
xy_cols = c("X", "Y"),
cutoff = 10)
ggplot() +
inlabru::gg(mesh$mesh) +
coord_fixed() +
geom_point(aes(X, Y), data = dd, alpha = 0.2, size = 0.5) +
annotate("text", -Inf, Inf, label = paste("n knots =", mesh$mesh$n), hjust = -0.3, vjust = 3) +
labs(x = "Easting (km)", y = "Northing (km)")
ggsave(paste0(home, "/figures/supp/mesh_", i, ".pdf"), width = 17, height = 17, units = "cm")
# Saduria model
if(unique(dd$group) == "flounder") {
m_sad <- sdmTMB(saduria_rel_weight ~ 0 + fyear + fquarter + depth_sc + oxygen_sc +
small_cod_density_sc*density_saduria_sc +
flounder_density_sc*density_saduria_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "off",
time = "year")
print(i)
sanity(m_sad)
print(m_sad)
} else {
m_sad <- sdmTMB(saduria_rel_weight ~ 0 + fyear + fquarter + depth_sc + oxygen_sc +
small_cod_density_sc*density_saduria_sc +
flounder_density_sc*density_saduria_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "on",
time = "year")
print(i)
sanity(m_sad)
print(m_sad)
}
# Benthic model
if(unique(dd$group) %in% c("large cod", "small cod")) {
m_ben <- sdmTMB(benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc +
small_cod_density_sc*oxygen_sc + flounder_density_sc*oxygen_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "off", # spatial off
time = "year")
print(i)
sanity(m_ben)
print(m_ben)
} else {
m_ben <- sdmTMB(benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc +
small_cod_density_sc*oxygen_sc + flounder_density_sc*oxygen_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "on",
time = "year")
print(i)
sanity(m_ben)
print(m_ben)
}
# Spatial and spatiotemporal random effects
d_haul <- dd %>%
distinct(haul_id, .keep_all = TRUE)
preds_sad <- predict(m_sad, newdata = d_haul)
preds_ben <- predict(m_ben, newdata = d_haul)
random_sad[[i]] <- preds_sad
random_ben[[i]] <- preds_ben
# Residuals
samps <- sdmTMBextra::predict_mle_mcmc(m_sad, mcmc_iter = 201, mcmc_warmup = 200)
mcmc_res <- residuals(m_sad, type = "mle-mcmc", mcmc_samples = samps)
dd$res <- as.vector(mcmc_res)
res_sad[[i]] <- dd
samps <- sdmTMBextra::predict_mle_mcmc(m_ben, mcmc_iter = 201, mcmc_warmup = 200)
mcmc_res <- residuals(m_ben, type = "mle-mcmc", mcmc_samples = samps)
dd$res <- as.vector(mcmc_res)
res_ben[[i]] <- dd
# Ranges
range_sad[[i]] <- tidy(m_sad, effects = "ran_pars") %>% filter(term == "range") %>% mutate(group = i, model = "saduria")
range_ben[[i]] <- tidy(m_ben, effects = "ran_pars") %>% filter(term == "range") %>% mutate(group = i, model = "benthos")
# Conditional effects: flounder
nd_flounder <- data.frame(expand_grid(
density_saduria_sc = c(quantile(d$density_saduria_sc, probs = 0.05),
quantile(d$density_saduria_sc, probs = 0.95)),
flounder_density_sc = seq(quantile(dd$flounder_density_sc, probs = 0.05),
quantile(dd$flounder_density_sc, probs = 0.95),
length.out = 50))) %>%
mutate(year = 2020,
fyear = as.factor(2020),
fquarter = as.factor(1),
oxygen_sc = 0,
depth_sc = 0,
small_cod_density_sc = 0,
fhaul_id = as.factor("2020_1_81")) # TODO: why do I need this when I specify re_form_iid!?
preds_flounder_sad <- predict(m_sad, newdata = nd_flounder, re_form = NA, re_form_iid = NA, se_fit = TRUE)
preds_flounder_ben <- predict(m_ben, newdata = nd_flounder, re_form = NA, re_form_iid = NA, se_fit = TRUE)
pred_flounder_sad[[i]] <- preds_flounder_sad %>% mutate(group = i, xvar = "flounder")
pred_flounder_ben[[i]] <- preds_flounder_ben %>% mutate(group = i, xvar = "flounder")
# Conditional effects: cod
nd_cod <- data.frame(expand_grid(
density_saduria_sc = c(quantile(d$density_saduria_sc, probs = 0.05),
quantile(d$density_saduria_sc, probs = 0.95)),
small_cod_density_sc = seq(quantile(dd$small_cod_density_sc, probs = 0.05),
quantile(dd$small_cod_density_sc, probs = 0.95),
length.out = 50))) %>%
mutate(year = 2020,
fyear = as.factor(2020),
fquarter = as.factor(1),
oxygen_sc = 0,
depth_sc = 0,
flounder_density_sc = 0,
fhaul_id = as.factor("2020_1_81")) # TODO: why do I need this when I specify re_form_iid!?
preds_cod_sad <- predict(m_sad, newdata = nd_cod, re_form = NA, re_form_iid = NA, se_fit = TRUE)
preds_cod_ben <- predict(m_ben, newdata = nd_cod, re_form = NA, re_form_iid = NA, se_fit = TRUE)
pred_cod_sad[[i]] <- preds_cod_sad %>% mutate(group = i, xvar = "cod")
pred_cod_ben[[i]] <- preds_cod_ben %>% mutate(group = i, xvar = "cod")
# Coefficients
coefs_sad <- bind_rows(tidy(m_sad, effects = "fixed", conf.int = TRUE)) %>%
mutate(species = "Cod (m)",
response = "Saduria",
sig = ifelse(estimate > 0 & conf.low > 0, "Y", "N"),
sig = ifelse(estimate < 0 & conf.high < 0, "Y", sig))
coefs_ben <- bind_rows(tidy(m_ben, effects = "fixed", conf.int = TRUE)) %>%
mutate(species = "Cod (m)",
response = "Saduria",
sig = ifelse(estimate > 0 & conf.low > 0, "Y", "N"),
sig = ifelse(estimate < 0 & conf.high < 0, "Y", sig))
coef_sad[[i]] <- coefs_sad %>% mutate(group = i)
coef_ben[[i]] <- coefs_ben %>% mutate(group = i)
}filter: removed 5,765 rows (63%), 3,345 rows remaining[1] "large cod"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: saduria_rel_weight ~ 0 + fyear + fquarter + depth_sc + oxygen_sc +
Formula: small_cod_density_sc * density_saduria_sc + flounder_density_sc *
Formula: density_saduria_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -8.36 1.02
fyear2016 -10.51 1.01
fyear2017 -10.77 0.95
fyear2018 -10.06 0.99
fyear2019 -11.32 1.14
fyear2020 -10.42 0.89
fyear2021 -10.29 0.90
fyear2022 -10.87 0.99
fquarter4 -0.46 0.29
depth_sc -0.96 0.26
oxygen_sc -0.02 0.22
small_cod_density_sc 0.43 0.37
density_saduria_sc 0.36 0.23
flounder_density_sc -0.36 0.25
small_cod_density_sc:density_saduria_sc 1.11 0.35
density_saduria_sc:flounder_density_sc -0.15 0.28
Dispersion parameter: 0.14
Tweedie p: 1.47
Matérn range: 49.49
Spatial SD: 2.50
Spatiotemporal IID SD: 1.74
ML criterion at convergence: -721.647
See ?tidy.sdmTMB to extract these values as a data frame.
[1] "large cod"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc + small_cod_density_sc *
Formula: oxygen_sc + flounder_density_sc * oxygen_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -6.56 0.30
fyear2016 -6.56 0.26
fyear2017 -6.76 0.22
fyear2018 -6.37 0.24
fyear2019 -7.25 0.29
fyear2020 -6.75 0.22
fyear2021 -6.44 0.21
fyear2022 -6.59 0.24
fquarter4 0.71 0.11
depth_sc -0.31 0.07
small_cod_density_sc 0.03 0.08
oxygen_sc 0.16 0.06
flounder_density_sc -0.22 0.06
small_cod_density_sc:oxygen_sc 0.10 0.09
oxygen_sc:flounder_density_sc 0.06 0.07
Dispersion parameter: 0.29
Tweedie p: 1.63
Matérn range: 28.79
Spatiotemporal IID SD: 0.92
ML criterion at convergence: -7117.386
See ?tidy.sdmTMB to extract these values as a data frame.distinct: removed 3,057 rows (91%), 288 rows remaining
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002759 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 27.59 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 17.021 seconds (Warm-up)
Chain 1: 0.051 seconds (Sampling)
Chain 1: 17.072 seconds (Total)
Chain 1:
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002573 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 25.73 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 10.564 seconds (Warm-up)
Chain 1: 0.024 seconds (Sampling)
Chain 1: 10.588 seconds (Total)
Chain 1: filter: removed 4 rows (80%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAfilter: removed 3 rows (75%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'small_cod_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'flounder_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA
mutate: new variable 'group' (character) with one unique value and 0% NAfilter: removed 7,196 rows (79%), 1,914 rows remaining[1] "small cod"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: saduria_rel_weight ~ 0 + fyear + fquarter + depth_sc + oxygen_sc +
Formula: small_cod_density_sc * density_saduria_sc + flounder_density_sc *
Formula: density_saduria_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -8.16 1.20
fyear2016 -9.30 1.14
fyear2017 -9.99 1.11
fyear2018 -9.72 1.28
fyear2019 -8.40 1.24
fyear2020 -9.12 1.09
fyear2021 -9.75 1.08
fyear2022 -9.14 1.10
fquarter4 -1.34 0.34
depth_sc -0.73 0.27
oxygen_sc -0.47 0.23
small_cod_density_sc 0.38 0.34
density_saduria_sc 0.55 0.24
flounder_density_sc -0.77 0.34
small_cod_density_sc:density_saduria_sc 0.25 0.30
density_saduria_sc:flounder_density_sc -0.15 0.39
Dispersion parameter: 0.19
Tweedie p: 1.50
Matérn range: 87.59
Spatial SD: 2.26
Spatiotemporal IID SD: 1.33
ML criterion at convergence: -591.883
See ?tidy.sdmTMB to extract these values as a data frame.
[1] "small cod"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc + small_cod_density_sc *
Formula: oxygen_sc + flounder_density_sc * oxygen_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -5.88 0.29
fyear2016 -5.85 0.25
fyear2017 -5.74 0.21
fyear2018 -5.65 0.24
fyear2019 -6.22 0.32
fyear2020 -5.73 0.21
fyear2021 -5.94 0.19
fyear2022 -5.66 0.21
fquarter4 0.42 0.10
depth_sc -0.34 0.07
small_cod_density_sc 0.07 0.07
oxygen_sc 0.16 0.06
flounder_density_sc -0.04 0.05
small_cod_density_sc:oxygen_sc -0.20 0.10
oxygen_sc:flounder_density_sc 0.02 0.07
Dispersion parameter: 0.11
Tweedie p: 1.54
Matérn range: 33.19
Spatiotemporal IID SD: 0.72
ML criterion at convergence: -5129.440
See ?tidy.sdmTMB to extract these values as a data frame.distinct: removed 1,647 rows (86%), 267 rows remaining
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002716 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 27.16 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 41.203 seconds (Warm-up)
Chain 1: 0.393 seconds (Sampling)
Chain 1: 41.596 seconds (Total)
Chain 1:
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002803 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 28.03 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 9.571 seconds (Warm-up)
Chain 1: 0.022 seconds (Sampling)
Chain 1: 9.593 seconds (Total)
Chain 1: filter: removed 4 rows (80%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAfilter: removed 3 rows (75%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'small_cod_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'flounder_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA
mutate: new variable 'group' (character) with one unique value and 0% NAfilter: removed 5,259 rows (58%), 3,851 rows remaining[1] "flounder"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: saduria_rel_weight ~ 0 + fyear + fquarter + depth_sc + oxygen_sc +
Formula: small_cod_density_sc * density_saduria_sc + flounder_density_sc *
Formula: density_saduria_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -8.33 1.60
fyear2016 -6.61 1.43
fyear2017 -9.24 1.42
fyear2018 -8.66 1.49
fyear2019 -8.84 1.53
fyear2020 -8.90 1.44
fyear2021 -13.05 1.79
fyear2022 -9.64 1.47
fquarter4 -0.46 0.20
depth_sc -0.43 0.18
oxygen_sc -0.02 0.21
small_cod_density_sc 0.10 0.26
density_saduria_sc -0.03 0.17
flounder_density_sc -0.80 0.23
small_cod_density_sc:density_saduria_sc -0.01 0.30
density_saduria_sc:flounder_density_sc -0.68 0.21
Dispersion parameter: 0.19
Tweedie p: 1.49
Matérn range: 111.89
Spatiotemporal IID SD: 3.08
ML criterion at convergence: -1764.896
See ?tidy.sdmTMB to extract these values as a data frame.
[1] "flounder"✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably largeSpatiotemporal model fit by ML ['sdmTMB']
Formula: benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc + small_cod_density_sc *
Formula: oxygen_sc + flounder_density_sc * oxygen_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -5.01 0.34
fyear2016 -5.17 0.28
fyear2017 -5.15 0.26
fyear2018 -5.59 0.30
fyear2019 -5.27 0.30
fyear2020 -4.68 0.25
fyear2021 -5.40 0.29
fyear2022 -5.02 0.27
fquarter4 0.11 0.09
depth_sc -0.22 0.09
small_cod_density_sc -0.03 0.06
oxygen_sc -0.02 0.06
flounder_density_sc -0.11 0.05
small_cod_density_sc:oxygen_sc -0.15 0.09
oxygen_sc:flounder_density_sc -0.01 0.06
Dispersion parameter: 0.18
Tweedie p: 1.51
Matérn range: 29.35
Spatial SD: 1.20
Spatiotemporal IID SD: 0.75
ML criterion at convergence: -5507.909
See ?tidy.sdmTMB to extract these values as a data frame.distinct: removed 3,590 rows (93%), 261 rows remaining
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002122 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 21.22 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 34.591 seconds (Warm-up)
Chain 1: 0.157 seconds (Sampling)
Chain 1: 34.748 seconds (Total)
Chain 1:
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002502 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 25.02 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 14.076 seconds (Warm-up)
Chain 1: 0.047 seconds (Sampling)
Chain 1: 14.123 seconds (Total)
Chain 1: filter: removed 3 rows (75%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAfilter: removed 4 rows (80%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'small_cod_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'oxygen_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'flounder_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'xvar' (character) with one unique value and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NAmutate: new variable 'group' (character) with one unique value and 0% NA
mutate: new variable 'group' (character) with one unique value and 0% NANow do a separate model for adult cod looking at total prey
dd <- filter(d, group == "large cod")filter: removed 5,765 rows (63%), 3,345 rows remaining mesh <- make_mesh(dd,
xy_cols = c("X", "Y"),
cutoff = 10)
# Total model
# NOTE: turning off spatial here due to convergence and AIC
m_tot <- sdmTMB(tot_rel_weight ~ 0 + fyear + fquarter + depth_sc + #(1|fhaul_id) +
large_cod_density_sc*oxygen_sc + flounder_density_sc*oxygen_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "off",
time = "year")
sanity(m_tot)✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably large print(m_tot)Spatiotemporal model fit by ML ['sdmTMB']
Formula: tot_rel_weight ~ 0 + fyear + fquarter + depth_sc + large_cod_density_sc *
Formula: oxygen_sc + flounder_density_sc * oxygen_sc
Mesh: mesh (isotropic covariance)
Time column: year
Data: dd
Family: tweedie(link = 'log')
coef.est coef.se
fyear2015 -4.20 0.20
fyear2016 -4.69 0.17
fyear2017 -4.60 0.14
fyear2018 -4.32 0.14
fyear2019 -4.75 0.21
fyear2020 -4.57 0.13
fyear2021 -4.39 0.13
fyear2022 -4.67 0.15
fquarter4 -0.19 0.09
depth_sc -0.04 0.05
large_cod_density_sc 0.13 0.06
oxygen_sc -0.03 0.05
flounder_density_sc -0.03 0.04
large_cod_density_sc:oxygen_sc 0.10 0.11
oxygen_sc:flounder_density_sc -0.07 0.05
Dispersion parameter: 0.62
Tweedie p: 1.71
Matérn range: 15.85
Spatiotemporal IID SD: 0.73
ML criterion at convergence: -7227.381
See ?tidy.sdmTMB to extract these values as a data frame. # Residuals
samps <- sdmTMBextra::predict_mle_mcmc(m_tot, mcmc_iter = 201, mcmc_warmup = 200)
SAMPLING FOR MODEL 'tmb_generic' NOW (CHAIN 1).
Chain 1:
Chain 1: Gradient evaluation took 0.002636 seconds
Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 26.36 seconds.
Chain 1: Adjust your expectations accordingly!
Chain 1:
Chain 1:
Chain 1: Iteration: 1 / 201 [ 0%] (Warmup)
Chain 1: Iteration: 20 / 201 [ 9%] (Warmup)
Chain 1: Iteration: 40 / 201 [ 19%] (Warmup)
Chain 1: Iteration: 60 / 201 [ 29%] (Warmup)
Chain 1: Iteration: 80 / 201 [ 39%] (Warmup)
Chain 1: Iteration: 100 / 201 [ 49%] (Warmup)
Chain 1: Iteration: 120 / 201 [ 59%] (Warmup)
Chain 1: Iteration: 140 / 201 [ 69%] (Warmup)
Chain 1: Iteration: 160 / 201 [ 79%] (Warmup)
Chain 1: Iteration: 180 / 201 [ 89%] (Warmup)
Chain 1: Iteration: 200 / 201 [ 99%] (Warmup)
Chain 1: Iteration: 201 / 201 [100%] (Sampling)
Chain 1:
Chain 1: Elapsed Time: 7.555 seconds (Warm-up)
Chain 1: 0.024 seconds (Sampling)
Chain 1: 7.579 seconds (Total)
Chain 1: mcmc_res <- residuals(m_tot, type = "mle-mcmc", mcmc_samples = samps)
dd$res <- as.vector(mcmc_res)
res_tot <- dd
# Range
range_tot <- tidy(m_tot, effects = "ran_pars") %>% filter(term == "range") %>% mutate(group = "large cod", model = "total")filter: removed 3 rows (75%), one row remainingmutate: new variable 'group' (character) with one unique value and 0% NA new variable 'model' (character) with one unique value and 0% NA # Spatial and spatiotemporal random effects
d_haul <- dd %>%
distinct(haul_id, .keep_all = TRUE)distinct: removed 3,057 rows (91%), 288 rows remaining preds_tot <- predict(m_tot, newdata = d_haul)
# Coefficients
coefs_tot <- bind_rows(tidy(m_tot, effects = "fixed", conf.int = TRUE)) %>%
mutate(species = "Cod (m)",
response = "Saduria",
sig = ifelse(estimate > 0 & conf.low > 0, "Y", "N"),
sig = ifelse(estimate < 0 & conf.high < 0, "Y", sig))mutate: new variable 'species' (character) with one unique value and 0% NA new variable 'response' (character) with one unique value and 0% NA new variable 'sig' (character) with 2 unique values and 0% NA coefs_tot <- coefs_tot %>% mutate(group = "large cod")mutate: new variable 'group' (character) with one unique value and 0% NAMake dataframes
coef_df <- bind_rows(bind_rows(coef_sad) %>% mutate(model = "Saduria"),
bind_rows(coef_ben) %>% mutate(model = "Benthos"))mutate: new variable 'model' (character) with one unique value and 0% NA
mutate: new variable 'model' (character) with one unique value and 0% NAcoef_df <- coef_df %>% bind_rows(coefs_tot %>% mutate(model = "Total"))mutate: new variable 'model' (character) with one unique value and 0% NApred_cod_df <- bind_rows(bind_rows(pred_cod_sad) %>% mutate(model = "Saduria"),
bind_rows(pred_cod_ben) %>% mutate(model = "Benthos"))mutate: new variable 'model' (character) with one unique value and 0% NA
mutate: new variable 'model' (character) with one unique value and 0% NApred_flounder_df <- bind_rows(bind_rows(pred_flounder_sad) %>% mutate(model = "Saduria"),
bind_rows(pred_flounder_ben) %>% mutate(model = "Benthos"))mutate: new variable 'model' (character) with one unique value and 0% NA
mutate: new variable 'model' (character) with one unique value and 0% NAres_df <- bind_rows(bind_rows(res_sad) %>% mutate(model = "Saduria"),
bind_rows(res_ben) %>% mutate(model = "Benthos"))mutate: new variable 'model' (character) with one unique value and 0% NA
mutate: new variable 'model' (character) with one unique value and 0% NAres_df <- res_df %>% bind_rows(res_tot %>% mutate(model = "Total"))mutate: new variable 'model' (character) with one unique value and 0% NArandom_df <- bind_rows(bind_rows(random_sad) %>% mutate(model = "Saduria"),
bind_rows(random_ben) %>% mutate(model = "Benthos"))mutate: new variable 'model' (character) with one unique value and 0% NA
mutate: new variable 'model' (character) with one unique value and 0% NArandom_df <- random_df %>% bind_rows(preds_tot %>% mutate(model = "Total"))mutate: new variable 'model' (character) with one unique value and 0% NArange_df <- bind_rows(range_tot, bind_rows(range_ben), bind_rows(range_sad))Plot spatial random effects
random_df <- random_df %>%
mutate(group = str_to_sentence(group))mutate: changed 1,920 values (100%) of 'group' (0 new NA)# Saduria
sad_omega <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Saduria" & !group == "Flounder"), aes(X*1000, Y*1000, color = omega_s), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ group) +
labs(color = "Spatial\nrandom effect") +
theme(axis.text.x = element_text(angle = 90),
legend.position = "right",
legend.direction = "vertical",
legend.key.width = unit(0.4, "cm"),
legend.key.height = unit(0.4, "cm"))filter: removed 1,365 rows (71%), 555 rows remainingggsave(paste0(home, "/figures/supp/omega_sad.pdf"), width = 17, height = 7, units = "cm")
# Now do benthos (only for flounder)
ben_omega <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Benthos" & group == "Flounder"), aes(X*1000, Y*1000, color = omega_s), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ group) +
labs(color = "Spatial\nrandom effect") +
theme(axis.text.x = element_text(angle = 90),
legend.position = "right",
legend.direction = "vertical",
legend.key.width = unit(0.4, "cm"),
legend.key.height = unit(0.4, "cm"))filter: removed 1,659 rows (86%), 261 rows remainingben_omega
ggsave(paste0(home, "/figures/supp/omega_ben.pdf"), width = 11, height = 11, units = "cm")Plot spatiotemporal random effects
# Saduria
sad_eps_sc <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Saduria" & group == "Small cod"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,653 rows (86%), 267 rows remainingsad_eps_sc
ggsave(paste0(home, "/figures/supp/epsilon_sad_small_cod.pdf"), width = 17, height = 17, units = "cm")
sad_eps_lc <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Saduria" & group == "Large cod"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,632 rows (85%), 288 rows remainingsad_eps_lc
ggsave(paste0(home, "/figures/supp/epsilon_sad_large_cod.pdf"), width = 17, height = 17, units = "cm")
sad_eps_f <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Saduria" & group == "Flounder"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,659 rows (86%), 261 rows remainingsad_eps_f
ggsave(paste0(home, "/figures/supp/epsilon_sad_flounder.pdf"), width = 17, height = 17, units = "cm")
# Benthos
ben_eps_sc <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Benthos" & group == "Small cod"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,653 rows (86%), 267 rows remainingben_eps_sc
ggsave(paste0(home, "/figures/supp/epsilon_ben_small_cod.pdf"), width = 17, height = 17, units = "cm")
ben_eps_lc <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Benthos" & group == "Large cod"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,632 rows (85%), 288 rows remainingben_eps_lc
ggsave(paste0(home, "/figures/supp/epsilon_ben_large_cod.pdf"), width = 17, height = 17, units = "cm")
ben_eps_f <- plot_map_fc +
geom_point(data = random_df %>% filter(model == "Benthos" & group == "Flounder"), aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))filter: removed 1,659 rows (86%), 261 rows remainingben_eps_f
ggsave(paste0(home, "/figures/supp/epsilon_ben_flounder.pdf"), width = 17, height = 17, units = "cm")
# Total
tot_eps <- plot_map_fc +
geom_point(data = preds_tot, aes(X*1000, Y*1000, color = epsilon_st), size = 0.9) +
scale_color_gradient2() +
facet_wrap(~ year) +
labs(color = "Spatiotemporal\nrandom effect") +
theme(legend.position = c(0.84, 0.16),
axis.text.x = element_text(angle = 90))
tot_eps
Plot range
pal <- brewer.pal(n = 8, name = "Dark2")[c(2, 7, 6)]
range_df %>%
mutate(group = str_to_sentence(group),
model = str_to_sentence(model)) %>%
ggplot(aes(model, estimate, color = group)) +
geom_point(size = 2) +
geom_hline(yintercept = 15, linetype = 2, alpha = 0.5) +
scale_color_manual(values = pal) +
labs(x = "Prey group", y = "Range (km)", color = "Group") +
theme(aspect.ratio = 1,
legend.position = c(0.86, 0.86)) mutate: changed 7 values (100%) of 'group' (0 new NA) changed 7 values (100%) of 'model' (0 new NA)
ggsave(paste0(home, "/figures/supp/ranges.pdf"), width = 11, height = 11, units = "cm")Plot residuals
# Plot residuals
res_df |>
mutate(group = str_to_title(group)) |>
ggplot(aes(sample = res)) +
stat_qq(size = 0.75, shape = 21, fill = NA) +
facet_grid(model ~ group) +
stat_qq_line() +
labs(y = "Sample Quantiles", x = "Theoretical Quantiles") +
theme(aspect.ratio = 1)mutate: changed 21,565 values (100%) of 'group' (0 new NA)
ggsave(paste0(home, "/figures/supp/qq_relative_prey_weight.pdf"), width = 17, height = 17, units = "cm")Plot coefficients
coef_df$term %>% unique() [1] "fyear2015"
[2] "fyear2016"
[3] "fyear2017"
[4] "fyear2018"
[5] "fyear2019"
[6] "fyear2020"
[7] "fyear2021"
[8] "fyear2022"
[9] "fquarter4"
[10] "depth_sc"
[11] "oxygen_sc"
[12] "small_cod_density_sc"
[13] "density_saduria_sc"
[14] "flounder_density_sc"
[15] "small_cod_density_sc:density_saduria_sc"
[16] "density_saduria_sc:flounder_density_sc"
[17] "small_cod_density_sc:oxygen_sc"
[18] "oxygen_sc:flounder_density_sc"
[19] "large_cod_density_sc"
[20] "large_cod_density_sc:oxygen_sc" pal <- brewer.pal(n = 8, name = "Dark2")[c(2, 7, 6)]
# Fix some names
coef_df2 <- coef_df %>%
filter(!grepl('year', term)) %>%
filter(!grepl('quarter', term)) %>%
mutate(term = str_remove_all(term, "_sc"),
term = str_remove_all(term, "density"),
term = str_replace_all(term, "_", ""),
term = str_replace_all(term, "geco", "ge co"),
term = str_replace_all(term, "llco", "ll co"),
term = str_replace(term, ":", " × "),
term = str_to_sentence(term),
group = str_to_sentence(group))filter: removed 56 rows (52%), 52 rows remainingfilter: removed 7 rows (13%), 45 rows remainingmutate: changed 45 values (100%) of 'term' (0 new NA) changed 45 values (100%) of 'group' (0 new NA)ggplot(coef_df2, aes(estimate, term, color = group, alpha = sig)) +
geom_stripped_rows(aes(y = term), inherit.aes = FALSE) +
facet_wrap2(~model, ncol = 2, scales = "free") +
geom_vline(xintercept = 0, linetype = 2, alpha = 0.5, color = "gray10", linewidth = 0.2) +
geom_point(position = position_dodge(width = 0.5), size = 1.5) +
geom_errorbar(aes(xmin = conf.low, xmax = conf.high), width = 0,
position = position_dodge(width = 0.5)) +
scale_alpha_manual(values = c(0.4, 1)) +
scale_color_manual(values = pal) +
labs(x = "", y = "Standardized coefficient", alpha = "95% CI crossing 0", color = "Group") +
guides(color = guide_legend(title.position = "top", title.hjust = 0.5),
alpha = guide_legend(title.position = "top", title.hjust = 0.5)) +
theme(legend.position = c(0.75, 0.2),
legend.direction = "vertical",
legend.box.spacing = unit(-3, "pt"),
legend.margin = margin(0, 0, 0, 0)) +
NULL
ggsave(paste0(home, "/figures/coefs.pdf"), width = 17, height = 13, units = "cm")Plot year and quarter coefficients
# Fix some names
coef_df3 <- coef_df %>%
filter(grepl('year', term)) %>%
mutate(term = str_remove_all(term, "fyear"),
group = str_to_sentence(group),
term = as.numeric(term))filter: removed 52 rows (48%), 56 rows remainingmutate: converted 'term' from character to double (0 new NA) changed 56 values (100%) of 'group' (0 new NA)ggplot(coef_df3, aes(term, estimate, color = group, fill = group)) +
facet_wrap(~model, scales = "free", ncol = 1) +
geom_line(position = position_dodge(width = 0.5)) +
geom_ribbon(aes(ymin = conf.low, ymax = conf.high), alpha = 0.2, color = NA) +
scale_color_manual(values = pal) +
scale_fill_manual(values = pal) +
labs(x = "Year", y = "Standardized coefficient", color = "") +
guides(color = guide_legend(title.position = "top", title.hjust = 0.5, ncol = 3),
fill = "none") +
theme(legend.position = c(0.5, 0.99),
legend.direction = "vertical",
legend.box.spacing = unit(-3, "pt"),
legend.margin = margin(0, 0, 0, 0),
strip.text.x.top = element_text(angle = 0, hjust = 0)) +
NULL
ggsave(paste0(home, "/figures/supp/coefs_year.pdf"), width = 11, height = 21, units = "cm")
# Now do quarter
coef_df5 <- coef_df %>%
filter(term %in% c("fquarter4")) %>%
mutate(group = str_to_sentence(group))filter: removed 101 rows (94%), 7 rows remainingmutate: changed 7 values (100%) of 'group' (0 new NA)ggplot(coef_df5, aes(estimate, model, color = group, alpha = sig)) +
geom_vline(xintercept = 0, linetype = 2, alpha = 0.5, color = "gray10", linewidth = 0.2) +
geom_point(position = position_dodge(width = 0.5), size = 1.5) +
geom_errorbar(aes(xmin = conf.low, xmax = conf.high), width = 0,
position = position_dodge(width = 0.5)) +
scale_alpha_manual(values = c(0.4, 1)) +
scale_color_manual(values = pal) +
labs(x = "", y = "Quarter 4 effect", alpha = "95% CI crossing 0", color = "Group") +
guides(color = guide_legend(title.position = "top", title.hjust = 0.5),
alpha = guide_legend(title.position = "top", title.hjust = 0.5)) +
theme(legend.position = "bottom",
legend.direction = "horizontal",
legend.box = "horizontal",
legend.box.spacing = unit(-3, "pt"),
legend.margin = margin(0, 0, 0, 0))
ggsave(paste0(home, "/figures/supp/coefs_quarter.pdf"), width = 17, height = 11, units = "cm")Conditional effects
# Which CI?
# https://www.calculator.net/confidence-interval-calculator.html
pred_df <- bind_rows(pred_cod_df, pred_flounder_df) %>%
mutate(group = str_to_sentence(group),
sad = ifelse(density_saduria_sc == min(density_saduria_sc), "Low", "High"))mutate: changed 1,200 values (100%) of 'group' (0 new NA) new variable 'sad' (character) with 2 unique values and 0% NA# 75% CI!!
# x small cod for different saduria (y = small cod)
# x flounder cod for different saduria (y = all)
ggplot(pred_df %>% filter(model == "Saduria" & xvar == "cod" & group == "Small cod"),
aes(small_cod_density_sc, exp(est), color = sad, fill = sad)) +
geom_ribbon(aes(ymin = exp(est - 1.036*est_se), ymax = exp(est + 1.036*est_se)),
alpha = 0.3, color = NA) +
geom_line() +
#coord_cartesian(ylim = c(0, 0.0001)) +
facet_wrap(~group, scales = "free") +
scale_color_brewer(palette = "Dark2") +
scale_fill_brewer(palette = "Dark2") +
labs(x = "Small cod density", y = "Relative saduria weight",
color = "Saduria", fill = "Saduria") +
theme(legend.position = c(0.88, 0.88))filter: removed 1,100 rows (92%), 100 rows remaining
#ggsave(paste0(home, "/figures/conditional_saduria_scod.pdf"), width = 11, height = 11, units = "cm")
ggplot(pred_df %>% filter(model == "Saduria" & xvar == "flounder" & !group == "Large cod"),
aes(flounder_density_sc, exp(est), color = sad, fill = sad)) +
geom_ribbon(aes(ymin = exp(est - 1.150*est_se), ymax = exp(est + 1.150*est_se)),
alpha = 0.3, color = NA) +
geom_line() +
facet_wrap(~group, scales = "free", ncol = 2) +
scale_color_brewer(palette = "Dark2") +
scale_fill_brewer(palette = "Dark2") +
coord_cartesian(xlim = c(-0.5, 0.5)) +
labs(x = "Flounder density", y = "Relative saduria weight",
color = "Saduria", fill = "Saduria") +
theme(legend.position = c(0.95, 0.88),
strip.text.x.top = element_text(angle = 0, hjust = 0))filter: removed 1,000 rows (83%), 200 rows remaining
ggsave(paste0(home, "/figures/conditional_saduria_flounder.pdf"), width = 17, height = 9, units = "cm")
# Conditional effects of oxygen and flounder for the benthos model
nd_flounder_oxy <- data.frame(expand_grid(
oxygen_sc = c(quantile(d$oxygen_sc, probs = 0.05),
quantile(d$oxygen_sc, probs = 0.95)),
flounder_density_sc = seq(quantile(dd$flounder_density_sc, probs = 0.05),
quantile(dd$flounder_density_sc, probs = 0.95),
length.out = 50))) %>%
mutate(year = 2020,
fyear = as.factor(2020),
fquarter = as.factor(1),
density_saduria_sc = 0,
depth_sc = 0,
small_cod_density_sc = 0,
fhaul_id = as.factor("2020_1_81")) # TODO: why do I need this when I specify re_form_iid!?mutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'density_saduria_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'small_cod_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NA preds_flounder_oxy_ben <- predict(m_ben, newdata = nd_flounder_oxy, re_form = NA, re_form_iid = NA, se_fit = TRUE)
ggplot(preds_flounder_oxy_ben, aes(flounder_density_sc, exp(est), color = factor(round(oxygen_sc)), fill = factor(round(oxygen_sc)))) +
geom_ribbon(aes(ymin = exp(est - 1.150*est_se), ymax = exp(est + 1.150*est_se)),
alpha = 0.3, color = NA) +
geom_line() +
#facet_wrap(~group, scales = "free", ncol = 2) +
scale_color_brewer(palette = "Dark2") +
scale_fill_brewer(palette = "Dark2") +
coord_cartesian(xlim = c(-0.5, 0.5)) +
labs(x = "Flounder density", y = "Relative saduria weight",
color = "Oxygen", fill = "Oxygen") +
theme(legend.position = c(0.95, 0.88),
strip.text.x.top = element_text(angle = 0, hjust = 0))
Showing conditional effects of oxygen on small cod feeding on benthos
dd <- filter(d, group == "small cod")filter: removed 7,196 rows (79%), 1,914 rows remaining mesh <- make_mesh(dd,
xy_cols = c("X", "Y"),
cutoff = 10)
# Benthic model
m_ben <- sdmTMB(benthic_rel_weight ~ 0 + fyear + fquarter + depth_sc + #(1|fhaul_id) +
small_cod_density_sc*oxygen_sc + flounder_density_sc*oxygen_sc,
data = dd,
mesh = mesh,
family = tweedie(),
spatiotemporal = "IID",
spatial = "off",
time = "year")
sanity(m_ben)✔ Non-linear minimizer suggests successful convergence✔ Hessian matrix is positive definite✔ No extreme or very small eigenvalues detected✔ No gradients with respect to fixed effects are >= 0.001✔ No fixed-effect standard errors are NA✔ No standard errors look unreasonably large✔ No sigma parameters are < 0.01✔ No sigma parameters are > 100✔ Range parameter doesn't look unreasonably large nd <- data.frame(oxygen = seq(quantile(d$oxygen, probs = 0.05), quantile(d$oxygen, probs = 0.95),
length.out = 50)) %>%
mutate(year = 2020,
fyear = as.factor(2020),
fquarter = as.factor(1),
density_saduria_sc = 0,
flounder_density_sc = 0,
depth_sc = 0,
small_cod_density_sc = 0,
fhaul_id = as.factor("2020_1_81")) %>%
mutate(oxygen_sc = (oxygen - mean(d$oxygen)) / sd(d$oxygen)) mutate: new variable 'year' (double) with one unique value and 0% NA new variable 'fyear' (factor) with one unique value and 0% NA new variable 'fquarter' (factor) with one unique value and 0% NA new variable 'density_saduria_sc' (double) with one unique value and 0% NA new variable 'flounder_density_sc' (double) with one unique value and 0% NA new variable 'depth_sc' (double) with one unique value and 0% NA new variable 'small_cod_density_sc' (double) with one unique value and 0% NA new variable 'fhaul_id' (factor) with one unique value and 0% NAmutate: new variable 'oxygen_sc' (double) with 50 unique values and 0% NA p <- predict(m_ben, newdata = nd, re_form = NA, re_form_iid = NA, se_fit = TRUE)
ggplot(p, aes(oxygen, exp(est))) +
geom_line() +
theme_sleek(base_size = 14) +
geom_hline(yintercept = 0.0025, col = "red") +
geom_hline(yintercept = 0.004, col = "red") +
geom_vline(xintercept = 4, col = "red") +
geom_vline(xintercept = 8, col = "red") +
NULL